Why do I have to convert "uint8" into "float32"

the whole math for neural networks is continuous, not discrete, and this is best approximated with floating point numbers.

We should perform a lot of data augmentation  and training as well in CNN. This will lead to faster training experience